CHANGES IN
TSINTERMITTENT VERSION 2.0 (08 NOV 2026)
- new: Empirical quantiles of the demand over the lead time for all
methods, using the input probs. When probs is provided, h is understood
as the lead time, and the quantiles are calculated from the cumulative
forecast errors over h periods from all in-sample origins. The output
element lt contains the quantiles, the safety stock, and the cumulative
errors with their origins. The calculation is also exported as
lt.dist.
- new: New cost functions msce and mace, on the cumulative errors over
h periods, for crost, crost.ma, tsb, and sexsm.
- new: Outputs of crost, crost.ma, tsb, sexsm, and imapa are now
objects of class tsint (imapa also of class imapa), with print, summary,
plot, coef, fitted, and forecast methods. fitted and forecast return ts
objects when the input is a ts object; the list elements frc.in and
frc.out remain numeric vectors. idclass returns an object of class
idclass with print and plot methods. Existing list elements are
unchanged; the information used by the methods is appended in a new
element, fit.info.
- new: New forecast method (using the forecast generic of the generics
package), which gives point forecasts for any horizon without optimising
the model again, optionally applying it to new data. lt.dist accepts a
fitted model, to calculate quantiles for any lead time and probabilities
without refitting.
- new: New input model in crost, crost.ma, tsb, and sexsm to apply a
fitted model to new data without optimisation. With init.opt=TRUE, only
the initial values are optimised. imapa accepts a fitted model in
model.fit.
- new: Plots moved to plot methods, which accept graphical arguments,
e.g. main and ylim. New lead time view of the forecasts. outplot also
accepts the name of a view, e.g. outplot=“leadtime”.
- new: imapa: new input cl for parallel processing with an existing
cluster. paral=1 still looks for a cluster named cl in the global
environment if cl is not provided.
- new: Added a vignette and tests.
- imapa no longer depends on the MAPA package. Requires R >=
4.1.0.
- imapa: the default maximumAL is the larger of the maximum
inter-demand interval of the series and its sampling frequency (for ts
objects), instead of half the length of the series. Results of imapa
without maximumAL differ from earlier versions.
- imapa: points in the model selection plot have a black outline.
- imapa: parallel processing checks that the workers can load the same
version of the package, and gives an informative error otherwise, e.g.,
when the package is loaded with devtools::load_all().
- Documentation and NAMESPACE are now generated with roxygen2 from the
R files, with more examples. Added package level help page
(?tsintermittent).
- Input checks on type, cost, init, and outplot, with informative
errors. type=“crost” is accepted as an alias for “croston”.
- opt.on is deprecated and will be removed in a future version.
- bugfix: crost, tsb, sexsm: the length of a numeric init was checked
incorrectly. crost and tsb now give a clear error if init has fewer than
two values.
- bugfix: tsb: failed when a single smoothing parameter was given and
init.opt=TRUE. tsb now gives a clear error.
- bugfix: crost.ma: when the grid search had tied minima, the orders
of different minima were combined.
- bugfix: data.frc: leftover browser() call and wrong type with
method=“auto”. Series that cannot be classified are now forecast with
SES.
- bugfix: data.frc: failed when h=1.
- bugfix: idclass: failed for a single ts series. Series with fewer
than two non-zero values are now left out of all classes with a
warning.
- bugfix: imapa: returned matrices when a single aggregation level was
used, and failed when h=0.
CHANGES IN
TSINTERMITTENT VERSION 1.10 (15 JUL 2022)
- Updated links.
- Updated DESCRIPTION.
- bugfix: Fixed checks on classes.
- bugfix: crost, crost.ma: fixed bug when checking number of
parameters to optimise. Bug did not cause issues before.
CHANGES IN
TSINTERMITTENT VERSION 1.9 (10 MAR 2016)
- new: crost, crost.ma: provide vectors of non-zero demand and
intervals.
- crost.ma: changed the variable name of demand rate to “cc”.
CHANGES IN
TSINTERMITTENT VERSION 1.8 (23 SEP 2015)
- new: Added “auto” method in data.frc. This selects and uses Croston,
SBA and SES automatically for each time series, based on PKa
classification (idclass function).
- new: Added check on crost.decomp and idclass when input is a data
frame.
- new: Added warning on functions that handle single time series when
data frames with more than a single column are used as inputs.
- imapa: changed classification to PKa instead of PK with prespecified
alpha. Note that data.frc with method=“auto” and imapa with maximumAL=1
give the same results, but data.frc is faster. Use imapa when temporal
aggregation is desirable.
CHANGES IN
TSINTERMITTENT VERSION 1.7 (17 SEP 2015)
- new: Added function data.frc. This is a wrapper to forecast data
frames with a single command.
- new: Added option to remove NA values in all forecasting
methods.
- simID now outputs different series as columns in a data frame.
CHANGES IN
TSINTERMITTENT VERSION 1.6 (17 FEB 2015)
- new: Added separate function crost.decomp for Croston
decomposition.
CHANGES IN
TSINTERMITTENT VERSION 1.5 (20 NOV 2014)
- bugfix: crost.ma: corrected optimisation error.
CHANGES IN
TSINTERMITTENT VERSION 1.4 (03 OCT 2014)
- new: Added more time series checks for crost and imapa
functions.
- imapa: revised default maximumAL to be floor(number of
observations/2).
- simID: changed default CV^2 to 0.5.
CHANGES IN
TSINTERMITTENT VERSION 1.3 (04 AUG 2014)
- new: Added option to optimise only initial values without parameters
in crost, tsb and sexsm functions.
- new: Added init.opt option for imapa in the same way as it is for
crost and tsb functions. This does not affect SES optimisation.
- new: Added option to use pre-fitted imapa models.
- bugfix: Corrected major optimisation bug - issue with opt.on in all
crost, tsb and sexsm functions.
- bugfix: imapa: corrected bug when maximumAL was very high.
- bugfix: imapa: corrected bugs in the legend of plots.
CHANGES IN
TSINTERMITTENT VERSION 1.2 (16 JUL 2014)
- new: Introduced imapa function for intermittent demand series.
- Added MAPA package dependency for temporal aggregation
functions.
- Added parallel package dependency for imapa parallelisation.
CHANGES IN
TSINTERMITTENT VERSION 1.1 (03 JUL 2014)
- new: Added SES function (sexsm) with control over parameter and
initialisation.
- Added URL in description.
- Renamed data and data2 to ts.data1 and ts.data2.
- Forced type input to lowercase.
CHANGES IN
TSINTERMITTENT VERSION 1.0 (23 JUN 2014)